Comparative Analysis of the Predictive Ability of the Extended Cox Model

A study presents a comparative analysis of the extended Cox model with modern survival analysis methods. Purpose: to evaluate the predictive abilities of the extended Cox model in comparison with the current survival analysis models and techniques. To achieve this goal, machine learning methods (survival random forest, gradient boosting, support vector machines) and classical statistical approaches (Weibull, log-logistic, and log-normal models) were used. Methods: analyzing three datasets, that is prostate cancer patients, data on their recurrent admissions, and breast cancer patients. Results: To demonstrate that the extended Cox model outperforms or is comparable in accuracy to modern machine learning methods while maintaining high interpretability. Practical significance: the applicability of the extended Cox model in medicine, social sciences, and other fields where both prediction accuracy and understanding of the factors affecting the risk of an event are crucial. The scientific novelty of this study lies in conducting a first comparative analysis of the extended Cox model with other survival analysis methods opening new opportunities for improving and adapting the model in future research. The study will be of great importance for the development of survival analysis methods and their application in practical tasks contributing to increased prediction accuracy and improved interpretability of results.